User Guide for the TIMSS International Database.pdf - TIMSS and ...
User Guide for the TIMSS International Database.pdf - TIMSS and ...
User Guide for the TIMSS International Database.pdf - TIMSS and ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
C H A P T E R 5 S C A L I N G<br />
Likewise, <strong>the</strong> scoring function of response k to item i is a scalar, b ik, in <strong>the</strong> <strong>for</strong>mer, whereas it is<br />
a D by 1 column vector, b ik, in <strong>the</strong> latter.<br />
5.4 The Population Model<br />
The item response model is a conditional model in <strong>the</strong> sense that it describes <strong>the</strong> process of<br />
generating item responses conditional on <strong>the</strong> latent variable, q. The complete definition of <strong>the</strong><br />
<strong>TIMSS</strong> model, <strong>the</strong>re<strong>for</strong>e, requires <strong>the</strong> specification of a density, fq( qa ; ), <strong>for</strong> <strong>the</strong> latent<br />
variable, q. We use a to symbolize a set of parameters that characterize <strong>the</strong> distribution of q.<br />
The most common practice when specifying unidimensional marginal item response models<br />
is to assume that <strong>the</strong> students have been sampled from a normal population with mean m <strong>and</strong><br />
variance s 2 . That is:<br />
( )<br />
fqf qa ; ; , exp<br />
( ) º 2<br />
q ( qms)<br />
=<br />
1<br />
2<br />
2ps<br />
2<br />
é q - m ù<br />
ê-<br />
2 ú<br />
ëê<br />
2s<br />
ûú<br />
(8)<br />
or equivalently q = m+E<br />
(9)<br />
where<br />
2<br />
E ~ N(<br />
0,<br />
s ).<br />
Adams, Wilson, <strong>and</strong> Wu (1997) discuss how a natural extension of (8) is to replace <strong>the</strong> mean,<br />
T m, with <strong>the</strong> regression model, Yn b , where Y is a vector of u fixed <strong>and</strong> known values <strong>for</strong><br />
n<br />
student n, <strong>and</strong> b is <strong>the</strong> corresponding vector of regression coefficients. For example, Yn could<br />
be constituted of student variables such as gender, socio-economic status, or major. Then <strong>the</strong><br />
population model <strong>for</strong> student n, becomes<br />
T<br />
qn = Ynb+ En<br />
, (10)<br />
where we assume that <strong>the</strong> E are independently <strong>and</strong> identically normally distributed with mean<br />
n<br />
zero <strong>and</strong> variance s 2 so that (10) is equivalent to<br />
T T<br />
T<br />
fq qn; Yn, , s ps exp qn — Yn qn<br />
— Yn<br />
s b<br />
2 2 -12<br />
1<br />
( ) = 2<br />
é<br />
( ) - 2 ( b) ( b<br />
ù ) , (11)<br />
ëê 2<br />
ûú<br />
T 2<br />
a normal distribution with mean Yn b <strong>and</strong> variance s . If (11) is used as <strong>the</strong> population model<br />
<strong>the</strong>n <strong>the</strong> parameters to be estimated are b, s 2 , <strong>and</strong> x.<br />
The <strong>TIMSS</strong> scaling model takes <strong>the</strong> generalization one step fur<strong>the</strong>r by applying it to <strong>the</strong><br />
vector valued q ra<strong>the</strong>r than <strong>the</strong> scalar valued q resulting in <strong>the</strong> multivariate population model<br />
-d<br />
2 -1<br />
T -1<br />
fq( qn; Wn, g, S) = S exp<br />
é 1<br />
( 2p<br />
2 ) - ( qn — gWn) S ( qn — gW<br />
ù<br />
n)<br />
, (12)<br />
ëê 2<br />
ûú<br />
where g is a u ´ d matrix of regression coefficients, S is a d ´ d variance-covariance matrix<br />
<strong>and</strong> Wn is a u ´ 1 vector of fixed variables. If (12) is used as <strong>the</strong> population model <strong>the</strong>n <strong>the</strong><br />
parameters to be estimated are g, S, <strong>and</strong> x.<br />
In <strong>TIMSS</strong> we refer to <strong>the</strong> W n variables as conditioning variables.<br />
5 - 4 T I M S S D A T A B A S E U S E R G U I D E